Saliency detection based on the fusion of spatial and frequency domain analysis

Saliency detection is a challenging direction in the field of computer vision. To get the more accurate salient target, we propose a visual saliency detection method based on the fusion of spatial and frequency domain analysis. Firstly, the image is smoothed by a Gaussian filter and then segmented by the superpixel segmentation of SLIC. The spatial saliency map is calculated by the color distance. Secondly, the saliency map of frequency domain is obtained by hypercomplex Fourier transform. Finally, the point multiplication operation is adopted to merge the spatial and the frequency domain saliency map. Compared with the others methods, the model we proposed can well retain the complete contour, highlight the target and suppress the background^1.

I. Introduction

Visual attention mechanism plays an important regulatory role in the human visual system, which enables the human eye to acquire important information, transfer to and maintain the attention to this information efficiently. Saliency detection is the most important initial step in visual attention mechanism. It can quickly extract the regions of human visual interest from environmental background. Image saliency detection is widely used in image segmentation, target detection, image retrieval, image compression and other fields. Saliency detection is a low-level visual processing task in human visual system and also a preprocessed process. Effective saliency detection is of great significance and wide applicable value for computer vision tasks of higher level such as image analysis and recognition and visual application [1]–[2]. After more than 20 years of development, the saliency detection has obtained some achievements, but compared with the sophisticated human visual system, the present saliency detection model is still far behind, the accuracy is low and the detection efficiency needs to be improved, and has a great promotion space. Saliency detection will continue to be a hot research topic in computer vision and artificial intelligence in a long period of time in the future.